RE: https://neuromatch.social/@jonny/116331940556649057

this whole thread is incredible but I am absolutely screaming at \u2014

two hundred dollars per calendar month
apologies to anyone who already read me saying this on bsky but: for the price of claude code, I could pay off my mortgage 3-4 years earlier. so as a notional labour saving device, it is competing with "four years of additional uninterupted leisure time", which I would say is a pretty high bar to clear
the idea that I, a pretty capable computer programmer, should pay *four years of my life* to people who make 10x as much money as me to ship the worst dreck I have ever laid eyes on, is just profoundly insulting, on top of all the other harm this shit causes
I'm actually pleased I've spent several years very slowly writing a small program by actually learning a lot about what it needs to do and thinking about its design and painstakingly validating it. it beats the pants off putting my name on this garbage
having seen this, it's not at all surprising that the software is as slow as it is, that its output is as bad as it is, or that it costs so much to run. this is not work you produce if you have thought about how anything should work for more than 5 minutes
everyone I have ever worked with is leagues better at writing code than this, what are we doing here
until now I would have said that LLMs produce median-quality work, not high quality, kind of by definition. but this isn't even median quality, it is way below that
@jcoglan given that they *definitionally* have median-quality work as their input, the fact they output something worse than that is both surprising and obvious at the same time
@jcoglan Or maybe the median quality is way below than what you think 😁

@jcoglan Is it perhaps that it’s median-quality, but all done in isolation?

So each individual change is _fine_, but when you look at it in aggregate it’s worse because none of the changes take the others into account? Hence duplication, missed abstractions, etc.

@jcoglan The general view is that most code doesn't have to be good at all, it just needs to be written.

@buherator @jcoglan

They all say that.

But when they find bugs or need changes, thay don't understand how things "suddenly" got so bad! 🙄

@jcoglan People I know who use it say its outputs have become pretty good.

(I'm not touching that god damn thing unless I am forced to.)

@datarama I have tried it since the time they said it became really good and I do not agree with them
@jcoglan (In fairness, they've said that roughly every six months for the last three years. :P )
@jcoglan il leave own judgement out here. But if i were to disagree with people like Greg Kroa Hartmann, Thomas Ptacek and Nicholas Carlini at the same time that would seriously make me uncomfortable about my judgement
@jcoglan and the people who prompted that code get millions a year
@jcoglan that is before you take into account (per Forbes and other data I've seen first hand): "heavy Claude Code Max 20 user could rack up $5,000/month in API-equivalent usage"